When a new technology such as artificial intelligence creates new kinds of value, the traditional financial metrics (revenue, profit, return on investment) often fail to capture progress in the early years.
Instead, industries invent intermediate operating metrics: proxies that signal whether the new model is working before the business model is fully proven. Sometimes it works; sometimes it doesn't.
Lots of dot-com firms touted "eyeballs" as a measure of attention. Many competitive telecom firms used metrics such as "access line equivalents" (taking total bandwidth and breaking it into voice grade "line" equivalents) as an example of potential revenue upside.
These metrics usually measure one of three things:
Adoption (how many people use it)
Engagement or usage intensity
Network growth or installed base.
Stage | What Firms Measure |
Early technology adoption | Installed base, users, traffic |
Network growth | Engagement, interactions, ecosystem size |
Monetization phase | Revenue per user, margins |
Mature industry | Standard financial metrics |
In the computing business, there are many examples.
Technology Wave | Era | Early Operating Metric | What It Measured | Later Financial Metric That Replaced It | Firms |
Personal computers | 1980s | Installed PC base | Growth of computing platform | Software and hardware revenue | Microsoft, Apple |
Dial-up Internet | Early 1990s | Subscribers / online accounts | Growth of consumer internet access | ARPU and subscription revenue | America Online |
Web portals | Late 1990s | Page views | Traffic volume and advertising potential | Ad revenue per user | Yahoo |
Dot-com era websites | 1998–2001 | “Eyeballs” (unique visitors) | Audience reach | Advertising revenue | Netscape ecosystem sites |
Telecom data services | 1990s–2000s | Access Line Equivalents (ALEs) | Aggregate network demand | ARPU and service revenue | telecom carriers |
Search engines | Early 2000s | Queries per day | Demand for information retrieval | Revenue per search / ad revenue | Google |
Social media | 2005–2015 | Monthly Active Users (MAU) | Network size and engagement | Ad revenue per user | Meta Platforms |
Cloud computing | 2010s | Compute instances / workloads | Adoption of cloud infrastructure | Revenue growth and margin | Amazon Web Services |
SaaS software | 2010s | Annual Recurring Revenue (ARR) | Predictable subscription base | Free cash flow and margin | Salesforce |
Sharing economy | 2010s | Gross bookings / rides | Platform usage volume | Take rate and net revenue | Uber |
Streaming video | 2010s | Subscribers | Platform scale | ARPU and operating margin | Netflix |
Cryptocurrency | 2015–2022 | Wallets, hash rate, total value locked | Network security and participation | Transaction fees and financial services revenue | Coinbase ecosystem |
Generative AI | 2023–present | Tokens processed / active developers / API calls | Real workload demand | Revenue per model usage | OpenAI |
Many could note a similar pattern for AI. New metrics emerge because we cannot typically measure early impact using traditional financial measures:
Monetization lags adoption
Network effects require scale first
Investors need forward-looking signals, so usage metrics answer that question before profits exist.
Phase | Typical Metric |
Technology novelty | Install base |
Early growth | Users or traffic |
Platform stage | Engagement |
Business model maturity | Revenue per user |
Mature industry | Profitability |
The AI economy therefore creates new metrics in the interim:
Tokens processed
Active developers
Inference workload
Model training compute.
These resemble earlier indicators in the early internet era such as:
Eventually the industry will likely shift to measures more closely tied directly to firm profits and revenues:
Eventually, we’ll learn which operating metrics actually have higher predictive value, and which have less.
During the dot-com bubble around the turn of the century, some metrics turned out to have near-zero predictive value.
Company | Metric Highlighted | What the Metric Measured | Why It Was Misleading |
Pets.com | Website traffic / brand awareness | Consumer interest in online pet supplies | Traffic did not translate into profitable orders because shipping costs exceeded margins |
Webvan | Number of cities launched | Geographic expansion of grocery delivery infrastructure | Massive capital spending occurred before proving unit economics |
eToys | Revenue growth rate | Rapid expansion of online toy sales | Sales were heavily subsidized by marketing and discounting |
TheGlobe.com | Registered users | Size of social community platform | Users were mostly non-paying and generated little revenue |
Boo.com | Site engagement and global launch presence | Interest in online fashion retail | Extremely expensive website technology created slow performance and high operating costs |
Excite | Page views | Web portal traffic volume | Advertising demand could not support the scale of infrastructure spending |
Lycos | Unique visitors | Audience size of web portal | Monetization per visitor was extremely low |
Broadcast.com | Streaming traffic and media partnerships | Growth of internet audio/video streaming | Technology and bandwidth costs exceeded realistic revenue models |
Priceline (early phase) | Gross travel bookings | Total value of transactions handled | Gross bookings overstated the company’s actual revenue capture |
Drkoop.com | Health site visitors | Consumer interest in medical information | Advertising revenue insufficient to support operations |
